Data Scientist (AI & Real Estate Valuation)

Job description

Role summary

Data scientist who owns the development, validation, and ongoing improvement of AI prediction models for current and future real estate valuations in the Netherlands. You strengthen the intelligence layer behind the Quality Realty Marketplace (QRM) and its resellable AI Native Marketplace Platform (ANMP) — turning Dutch property data (and realtor-uploaded history) into trustworthy valuations and forecasts for realtors, vendors, and clients.

Current engineering phase
Production valuations today use the Altum AI API integrated in the live WordPress plugin. Your models will augment and eventually replace that dependency over time — not on day one. Early months focus on data audit, Excel ingest design, baselines, and experimentation; heavier ETL (Airflow) and production ML serving ramp in the second half of the 6-month plan. Target accuracy: MAPE < 6–8% on hold-out data where feasible.

About Quality Realty (QR)

Quality Realty is an early-stage AI-native SaaS startup building the future of real estate transactions in the Netherlands. We are developing Quality Realty Marketplace (QRM) — a voice-first, multilingual platform that connects realtors, clients, and service vendors (painters, movers, stagers, energy auditors, etc.) throughout the entire home buying, selling, and rental process.

At its core is the AI Native Marketplace Platform (ANMP) — a fully resellable, white-label, multi-tenant marketplace operating system. QR is the first vertical (Dutch real estate), designed from day one to power other local-service marketplaces worldwide.

Key responsibilities

• Collect, explore, clean, and preprocess data from Dutch sources (CBS, Kadaster, BAG, PDOK, ECB/OECD macros, and realtor Excel uploads) to build robust training and validation datasets. Funda/Pararius ingestion only where legally approved (may be deferred for MVP).
• Perform feature engineering, including geospatial work (e.g. geopy/geopandas), temporal lags, economic adjustments (interest rates, inflation), and Dutch-specific factors (energy labels, postcode granularity, house age, WOZ influence).
• Design, train, and iterate on ML models for property valuation (XGBoost, Random Forest, LightGBM) and, where scoped, time-series forecasting (LSTM, Prophet, or hybrid approaches) targeting MAPE < 6–8% on hold-out and realistic test sets.
• Benchmark against the existing Altum integration so we can switch or A/B without regressing tenant-facing quality.
• Conduct rigorous evaluation: cross-validation, backtesting against historical Dutch market shifts, and monitoring for concept drift (e.g. post-2022 rate changes, energy transition).
• Apply explainability (SHAP, LIME) so product and voice flows can surface understandable drivers (e.g. location, energy label contributions).
• Integrate models into the Python backend (FastAPI) for batch and real-time inference, working with the Middle Fullstack Developer and Junior Backend Developer on reliable, low-latency serving.
• Automate retraining pipelines (Cloud Composer / Airflow) and contribute to the ETL layer so models receive fresh data.
• Analyze errors, market trends, and feedback; propose improvements (e.g. renovation scenarios, risk flags, personalized insights).
• Document models, experiments, feature importance, and data usage for GDPR, internal knowledge sharing, and future white-label use.
• Partner with the Tech Lead on model selection, architecture, and scalability; support data-science best practices across the team as it grows.

Your skills and experience

Required skills & experience

• 2–4 years professional experience as a Data Scientist or ML Engineer with at least one model used in production or serious pilot; 3+ years welcome for stronger candidates.
• Strong Python for data science (pandas, numpy, scikit-learn, XGBoost/LightGBM; TensorFlow/PyTorch or statsmodels as needed).
• Solid experience with regression and preferably time-series forecasting; geospatial skills (geopandas, geopy) a significant plus.
• Proven ability to clean and merge large, messy datasets, handle outliers, and engineer features that improve accuracy.
• Practical knowledge of MAPE, RMSE, MAE, hyperparameter tuning (Optuna, GridSearch), cross-validation, and explainable AI (SHAP, LIME).
• Comfortable with SQL (PostgreSQL/MySQL) and Git.
• Strong analytical mindset; ability to translate real estate and business problems into data-driven solutions.
• Eagerness to work in a small startup, iterate quickly, and take ownership of outcomes.
• Good English for documentation and collaboration with the CTO and Tech Lead.

Nice to have

• Dutch real estate domain knowledge (energy labels, WOZ, regional trends, post-2020 market shifts).
• Experience deploying ML (Docker, FastAPI, Cloud Run, drift monitoring).
• Airflow / Cloud Composer or data quality tools (Great Expectations).
• MLflow or similar experiment tracking.
• Multimodal or LLM-enhanced features (images, RAG) — if scoped later.
• Dutch language for reading source documentation.

Why you'll love working here

What we offer

• Opportunity to shape the valuation intelligence of a modern, voice-first real estate marketplace with real user impact.
• Direct collaboration with the Tech Lead and CTO on AI strategy for QR and future ANMP resellers.
• Path to expanded responsibility as the platform and data mature.
• Remote-friendly environment with flexible hours (HCMC-based core team).
• Competitive compensation for Vietnam market + performance bonuses (level aligned with part-time scope; details with Tech Lead / CTO).
• Hands-on work with XGBoost, Dutch open data, GCP, and production integration challenges.
• Part-time ~20 hours/week; preferred start July 2026 (wave 4 in hiring plan — may adjust slightly by hire date). Heavy ETL/ML production peaks September–November 2026 in the current roadmap.

Reporting line

This role reports to the Tech Lead and works closely with the Middle Fullstack Developer (WordPress + Python), Junior Backend Developer, BA, and UX Designer to ensure models are production-ready, explainable, and compliant with product messaging (e.g. “AI estimate — not an official appraisal” until legal/product sign-off).